RNA-seq data and corresponding clinical information of glioma in level 3 HTSEQ-FPKM format were downloaded from the TCGA database (Tomczak et al., 2015), converted to TPM (Transcripts Per Million reads) format, and converted into log2 format for further analysis. Statistical analysis was performed using R software v3.6.3, and statistical significance was set at p < 0.05.

U87 (human glioblastoma cell line) and SVG (human astrocyte cell line) were purchased from the Institute of Basic Medicine, China Medical College. Two kinds of cell lines were cultured in Dulbecco’s modified Eagle’s medium (DMEM) containing with 10% fetal bovine serum at 37°C under an atmosphere of 5% CO₂.

Glioma samples were collected from Department of Neurosurgery, PLA General Hospital. Seventeen paraffin-embedded samples (1 case was normal brain tissue, 2 cases were grade 2, 4 cases were grade 3 and 10 cases were grade 4) were used for immunohistochemistry staining. The patient was informed of the content of this research and signed an informed consent before the operation. The study was approved by the Institutional Research Ethics Committee of the PLA General Hospital (batch number: S2018-089-01).

We investigated KIF18A expression in different tumour types from the TCGA database, which combined various types of information to analyze the role of KIF18A in multifarious tumours. We also investigated the expression of KIF18A in glioma and its subtypes (low-grade glioma [LGG] and glioblastoma [GBM]). In addition, The Human Protein Atlas (HPA) database (Pontén et al., 2008) was utilised to analyze the available immunohistochemical images. The R software v3.6.3 was utilised to perform the statistical analysis, and visualisation was implemented using the ggplots2 package v3.3.3.

Box plots are generated to examine the correlations of KIF18A expression with diverse clinical characteristics (age, WHO grade, IDH status, 1p/19q codeletion, primary therapy outcome, and OS event) in patients. The R software v3.6.3 was utilised to perform the statistical analysis, and visualisation was implemented using the ggplots2 package v3.3.3.

Volcano plots and heat maps were used to visualise the results of the top DEGs. The R software v3.6.3 was utilised to perform the statistical analysis, and visualisation was implemented using the ggplots2 package v3.3.3. Additionally, the interaction network of KIF18A-relative biological function was analyzed using the Metascape database (Zhou et al., 2019).

To deepen our understanding of the landscape of the KIF18A-related immune microenvironment, the immune and stromal scores, ESTIMATE scores, and immune cell infiltration were calculated and compared between the KIF18A-high and KIF18A-low groups. Statistical analysis was performed using Estimate package v1.0.13. In addition, the enrichment score of 24 immune cells was evaluated based on ssGSEA algorithms (Hänzelmann et al., 2013) utilizing the GSVA package v1.34.0. This package was also used to perform correlation analysis between immune infiltration and KIF18A expression levels.

Kaplan-Meier plots were used to evaluate the relationship between KIF18A expression and OS of patients. In addition, a time-dependent receiver operating characteristic (ROC) curve was constructed for KIF18A expression in gliomas. The R software v3.6.3 was used to perform the statistical analysis, and visualisation was accomplished using the Survminer package v0.4.9 version. The multivariate Cox regression was performed in the hypothesis test, we used the R software v3.6.3 to perform the statistical analysis, and accomplished visualisation using the survival package v3.2-10.

The KIF18A associated gene-gene interaction network was established using the GeneMANIA database (WardeFarley et al., 2010). The KIF18A-related protein-protein interaction (PPI) network was established using the STRING online database (Szklarczyk et al., 2021). Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyzes were performed for the top 20 KIF18A-binding proteins. The R cluster Profiler package v3.14.3 was used for statistical analysis, and the ggplot2 package v3.3.3 was used for visualisation.

Tissues were fixed in 4% paraformaldehyde, embedded in paraffin, and cut into 4 μm slices. These slices were placed on slides and processed as previously described. The tissue slices were incubated with primary anti-KIF18A (1:1000, 19245-1-AP, Proteintech) antibody diluted with 1% goat serum (Balb, WE0320) in PBS overnight. Tissue slices were incubated with the secondary antibody for 1 h. Then, the ABC Horseradish Peroxidase kit (Vector Laboratories) was used to stain the slices and visualisation was accomplished using DAB (3,3′-diaminobenzidine). Nuclei were counterstained with haematoxylin. The immunohistochemical staining results were analyzed and scored by two pathologists who were blinded to the sources of the clinical samples. The intensity of staining was analyzed by the semiquantitative integration method.

Equal amounts of protein lysates (30 μg) of two kinds of cell line were separated on 10% SDS-PAGE gels and transferred into a PVDF membrane (0.45µm; Amersham Bioscience, Freiburg, Germany). After blocking with milk, the membrane was incubated with primary antibodies KIF18A (19245-1-AP, Proteintech, Wuhan, China) and GAPDH (10494-1-AP, Proteintech, Wuhan, China).

For bioinformatics analysis, the statistical significance between two groups was detected by using the Wilcoxon rank sum test, and the Kruskal-Wallis test and Dunn’s tests were utilised for comparison of multi-groups. The correlation between KIF18A expression and other immune-relevant genes was calculated and evaluated by Spearman’s correlation coefficient. All statistical analysis was performed using R software (version 3.6.3), and two-tail p < 0.05 was considered as of statistical significance.

The analysis of KIF18A expression in different tumour types from the TCGA database revealed that apart from acute myeloid leukemia (AML), kidney chromophobe (KICH), and testicular germ cell tumours (TGCT), KIF18A was highly expressed in nearly all the tumours when compared with normal tissue (Figure 1A). In addition, the expression of KIF18A in GBM, LGGs and all gliomas was significantly higher than that in normal brain tissue (Figures 1B–D ). We then analyzed KIF18A protein expression using the HPA database and showed the immunohistochemical staining results in Figures 1E–G. Taken together, glioma displays prominently higher KIF18A expression than that in normal brain tissue at both the mRNA and protein levels.

To further clarify the role of KIF18A in glioma, we analyzed the mRNA expression of KIF18A in different clinical subgroups. First, KIF18A mRNA expression levels were increased in patients over 60 years of age (Figure 2A). In addition, the expression level of KIF18A increased with an increase in tumour grade, and the highest expression level of KIF18A was found in grade 4, the highest degree of malignancy (Figure 2B). According to the IDH status, the expression of KIF18A was significantly higher in wild-type patients than in mutant patients (Figure 2C). Lower expression of KIF18A was found in the 1p/19q codeletion group compared to that in the non-codeletion group (Figure 2D). In addition, KIF18A expression was significantly lower in patients with partial response and complete response than in patients with other primary treatment outcomes (Figure 2E). In terms of OS, KIF18A mRNA expression levels in the dead group were higher than those in the surviving group (Figure 2F). In summary, KIF18A mRNA expression is related to the clinical characteristics of glioma, and its high expression might suggest a poor prognosis.

According to the threshold values of |log2 fold-change (FC)|>1.5 and adjusted p-value<0.05, 1233 DEGs were obtained, including 952 upregulated genes and 281 downregulated genes (Figure 3A). The heat maps showed the top 20 DEGs, including HOXC13, SAA2, HOXD13, HOXC11, and OTX2 (Figure 3B). In addition, functional enrichment analysis was conducted to identify the most relevant biological functions of these DEGs, which revealed that DEGs were mainly enriched in functions related to cell division, including cell cycle, DNA replication, microtubule cytoskeleton organization, and meiotic cell cycle (Figure 3C). To further capture the correlation between these enriched biological functions, we mapped them into a network plot using metascape online database, as shown in Figure 3D. And the specific information of important nodes related to cell division in the network was shown in detail in the Supplementary Table S1.

To further investigate the mechanism underlying differences in survival between the KIF18A-high and KIF18A-low expression groups, we analyzed the correlation between KIF18A expression level and tumour microenvironment. According to the TCGA database, patients with elevated KIF18A expression tended to have markedly higher immune score (Figure 4A), stromal score (Figure 4B), and estimate score (Figure 4C). Enrichment analysis of immune infiltrating cells was performed, and the results revealed that aDC, cytotoxic cells, eosinophilis, iDC, macrophages, neutrophils, CD56dim cells, NK cells, T cells, T helper cells, and Th2 cells were clearly increased in the microenvironment of the KIF18A-high group (Figure 4D). In addition, we studied the correlation between multifarious immune infiltration cells and KIF18A expression levels, and found that Th2 cells, macrophages, eosinophils, aDC, and neutrophils were significantly positively correlated with KIF18A expression. In contrast, pDC, NK CD56bright cells, TFH, CD8 T cells, and Tem were markedly negatively correlated with the expression of KIF18A (Figure 4E). In summary, the expression of KIF18A may reflect immune infiltration status.

As previously mentioned, KIF18A expression was significantly associated with the malignant characteristics of gliomas. We then investigated the prognostic value of the KIF18A expression level. According to Kaplan-Meier plots, patients with higher KIF18A expression had worse OS (Figure 5A). The time-dependent ROC curve showed that the expression level of KIF18A had a good predictive effect on glioma survival (Figure 5B). In Multivariate Cox regression analysis, the elevated expression of KIF18A was found to be an independent risk factor for poor prognosis in glioma patients (Figure 5C). In other words, the higher the expression of KIF18A, the worse the prognosis.

To further assess the function of KIF18A, we investigated the genes and proteins interacting with KIF18A. First, a gene-gene interaction network was automatically constructed to identify the top 20 relevant genes associated with KIF18A utilising GeneMANIA online database, including RRP7A, ANAPC4, WEE1, CDC20, and NDC80, and most of them were related to the cell cycle and cell division (Figure 6A). In addition, according to STRING online database, a protein-protein interaction network was constructed, and the top 20 proteins relevant to KIF18A were identified (Figure 6B). We further analyzed these KIF18A binding proteins and their corresponding biological functions (Figure 6C). According to GO and KEGG enrichment analysis (Figure 6D), the results revealed that the primary biological processes (BP) included sister chromatid segregation, mitotic nuclear division, and mitotic sister chromatid segregation. The cellular component (CC) was primarily enriched in the chromosome, centromeric region, microtubule, and spindle. The molecular function (MF) was mainly involved in tubulin binding, microtubule binding, and microtubule motor activity. KEGG pathway enrichment was mainly related to phagosomes, oocyte meiosis, and cell cycle.

To verify the above results obtained from multifarious databases, we collected seventeen paraffin-embedded samples from patients with glioma, and immunohistochemical tests were performed on these samples. The results suggested that patients with higher grade of glioma tend to have more KIF18A expression (Figures 7A, B). In addition, according to western blot, we found that compared with the SVG cell line, KIF18A was highly expressed in the U87 cell line (Figures 7C, D). In conclusion, the experiment results verified by clinical samples and cell lines was consistent with the data from bioinformatics.